[ NNSquad ] Re: Comcast announces tests of "protocol agnostic network management"

["David P. Reed" <dpreed@reed.com>]

Umm ... whence comes the idea that Comcast should decide among a 
*particular* end-user's traffic what traffic to degrade when it decides 
that a particular user is overloading the network?

That's the role of "DPI" proposed in this thread.

When the end user observes that his aggregate upstream capacity is 
reduced, it's perfectly reasonable for that end user to select which 
traffic goes through, at a box that the he/she owns and controls: i.e. a 
NAT router.

Even without NAT, a very simple solution that works is as follows: let 
your RTP packets (VoIP) continue to stream at a constant rate.  TCP will 
then see this as a reduction of available upstream capacity, and back off.

One can also tune VoIP to back off to a lower bitrate codec when 
congestion is experienced along the path to destination.

Thus, a simple design rule of not mucking with what you don't have to 
would argue strongly against using the big hammer of DPI against a 
trivial problem like dealing with variable upstream capacity (after all, 
that variability's what the Internet is *designed for*, and what VoIP 
and other things expect as Normal Operating Conditions).

Of course, those who are arguing the case presuming the answer to "is 
DPI needed" MUST be "yes", those folks will not be happy.  

Those folks include the vendors of DPI equipment - NebuAd, Phorm, 
Ellacoya, Sandvine - since the current market for DPI - NSA and friends 
- is a low-growth market; the folks at various ISPs looking for a new 
profit stream to be gained by monetizing user behavior measures and 
monetizing the "value" of transactions carried over their network, 
rather than the competitive value of bit transport.

Those folks may not include people like small ISPs - small ISPs can't 
afford DPI equipment, and can't do NebuAd-style ad-insertion, or for 
that matter, support billing services that monetize VoIP as a special 
value add service, blocking it otherwise.  Small ISPs (like some here) 
would prefer that their customers prioritize their own traffic in 
response to resource reductions, using standard endpoint-based 
congestion control, I would think.